CN114781243A

CN114781243A - ETA prediction and model training method, device, medium and product

Info

Publication number: CN114781243A
Application number: CN202210289342.4A
Authority: CN
Inventors: 代睿; 唐翠; 崔恒斌; 秦伟; 李伟; 甘杉林
Original assignee: Autonavi Software Co Ltd
Current assignee: Autonavi Software Co Ltd
Priority date: 2022-03-22
Filing date: 2022-03-22
Publication date: 2022-07-22

Abstract

The embodiment of the disclosure discloses an ETA prediction and model training method, equipment, a medium and a product, wherein the method comprises the steps of obtaining the road section type of each road section in a target route; calculating a predicted passing time value of each road section in future time slices after the departure time based on a road section grade prediction model corresponding to the road section type of each road section; determining the predicted entering time of each road section based on the predicted passing time value of each road section in each future time slice; accumulating the predicted passing time value of each road section at the predicted entering moment to obtain the predicted passing time of the road section level of the target route; and predicting the corresponding predicted arrival time of the target route based on the predicted traffic time of the section level of the target route and the relevant traffic characteristics of the target route by using a route level prediction model. The technical scheme can predict the predicted arrival time corresponding to the target route more accurately.

Description

ETA prediction and model training method, device, medium and product

Technical Field

The disclosure relates to the technical field of data processing, in particular to an ETA prediction and model training method, equipment, medium and product.

Background

ETA (Estimated Time of Arrival) is an estimate of the length of a planned route trip. Typically, the navigation application will also use an ETA prediction algorithm to calculate for the user the ETA required for the user to reach the destination through the corresponding navigation route when planning the navigation route for the user. For navigation application with high daily life, the ETA prediction algorithm has considerable calling frequency, so that extremely high requirements are provided for the accuracy and the calculation efficiency of the ETA prediction algorithm. The current ETA prediction algorithm usually directly uses a prediction model to predict the current transit time of each road section on a navigation route, and the ETA of the navigation route is obtained through accumulation. The current passing time of different road sections in the prediction algorithm is obtained by prediction through a unified prediction model, and actually, the result can be rapidly predicted through a simple model for a plurality of road sections with stable passing time, so that the unified complex prediction model is used for different road sections in the existing ETA prediction algorithm, the calculation resource is wasted, the calculation efficiency is low, the current passing time of each road section is predicted in the existing model, the specific time of entering the road section is not considered, and the ETA predicted in such a way is not accurate.

Disclosure of Invention

To solve the problems in the related art, embodiments of the present disclosure provide an ETA prediction and model training method, apparatus, medium, and product.

In a first aspect, an ETA prediction method is provided in the embodiments of the present disclosure.

Specifically, the ETA prediction method includes:

acquiring the section type of each section in the target route;

calculating a predicted passing time value of each future time slice of each road section after the departure time based on a road section grade prediction model corresponding to the road section type of each road section;

determining the predicted entering time of each road section based on the predicted passing time value of each road section in each future time slice;

accumulating the predicted passing time value of each road section in the corresponding target time slice to obtain the estimated passing time of the road section level of the target route, wherein the target time slice corresponding to each road section is the future time slice of the estimated entering time of each road section;

and predicting the corresponding predicted arrival time of the target route based on the predicted traffic time of the section level of the target route and the relevant traffic characteristics of the target route by using a route level prediction model.

In one possible implementation, the type of the road segment includes at least one of the following types: the method comprises the following steps of (1) low-frequency road sections, stable low-flow road sections, short-time sudden congestion road sections and long-time sudden congestion road sections; the step of calculating the predicted value of the passing time of each road section in each future time slice after the departure time based on the road section level prediction model corresponding to the road section type of each road section comprises the following steps:

when the type of the road section is a low-frequency road section, obtaining a predicted value of the passing time of the road section in each future time slice based on the historical average passing time of the road section;

when the type of the road section is a stable low-flow road section, obtaining a predicted passing time value of the road section in each future time slice by using a linear model corresponding to a scene where the road section is located and based on the passing road condition characteristics of the road section;

when the type of the road section is a short-time sudden congestion road section, obtaining a predicted value of the passing time of the road section in each future time slice by using a short-time prediction depth model and based on the short-time road condition characteristics of the road section;

and when the type of the road section is a long-term sudden congestion road section, obtaining the predicted value of the passing time of the road section in each future time slice by using a long-term prediction depth model and based on the long-term road condition characteristics of the road section.

In one possible implementation, the method further includes:

determining the scene of the road section based on at least one of the following characteristics of the road section: the road section is located in the area, the time type of the departure time and the road grade of the road section.

In one possible implementation, the link type includes at least one of the following types under a primary link type: the method comprises the following steps of (1) low-frequency road sections, stable low-flow road sections, short-time emergent congestion road sections and long-time emergent congestion road sections, wherein the type of the first-stage road section comprises a common road section or a road section divided into directions at an intersection;

when the road section type of the road section is a stable low-flow road section under the corresponding first-level road section type, obtaining a passing time predicted value of the road section in each future time slice by using a linear model corresponding to a scene where the road section is located based on the passing road condition characteristics of the road section under the corresponding first-level road section type, wherein the linear model is a model corresponding to the stable low-flow road section under the corresponding first-level road section type, and the passing road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time passing characteristics and historical average passing time sequences corresponding to the future time slices; the passing road condition characteristics of the road sections under the intersection direction-dividing road section types comprise at least one of the following characteristics: real-time traffic characteristics, historical average traffic time sequences corresponding to the future time slices, intersection types and steering action types;

when the type of the road section is a short-time sudden congestion road section under the corresponding first-level road section type, obtaining a predicted value of the passing time of the road section in each future time slice by using a short-time prediction depth model corresponding to the corresponding first-level road section type and based on the short-time road condition characteristics of the road section under the corresponding first-level road section type, wherein the short-time road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time average passing time sequence in a short term, historical average passing time sequence corresponding to each future time slice, and real-time traffic flow time sequence in the short term; the short-time road condition characteristics of the road section under the intersection diversion road section type comprise at least one of the following characteristics: real-time average passing time sequence in a short term, historical average passing time sequence corresponding to each future time slice, real-time traffic flow time sequence in the short term, intersection type and steering action type;

when the type of a road section of the road section is a long-term sudden congestion road section under a corresponding first-level road section type, obtaining a predicted value of the passing time of the road section in each future time slice by using a long-term prediction depth model corresponding to the corresponding first-level road section type based on the long-term road condition characteristics of the road section under the corresponding first-level road section type, wherein the long-term road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time average traffic time sequence in a long term, historical average traffic time sequence corresponding to each future time slice, and real-time traffic flow time sequence in the long term; the long-term road condition characteristics of the road section under the road section type divided by the intersection comprise at least one of the following characteristics: real-time average passing time sequence in a long term, historical average passing time sequence corresponding to each future time slice, real-time traffic flow time sequence in a long term, intersection type and steering action type.

In a possible implementation manner, the short-term prediction depth model or the long-term prediction depth model includes a first time-domain gated convolutional layer, a graph attention layer, a second time-domain gated convolutional layer, a third time-domain gated convolutional layer, and a multi-layer perceptron MLP layer, which are connected in sequence.

In one possible implementation, the method further includes:

training to obtain a linear model corresponding to a target scene based on first sample data, wherein the first sample data comprises traffic road condition characteristics of a first sample section at a first sample moment and real traffic time values of time slices after the first sample moment, and the first sample section is a stable low-flow section in the target scene;

training to obtain a corresponding short-term prediction depth model based on second sample data, wherein the second sample data comprise short-term road condition characteristics of a second sample road section at a second sample moment and real traffic time values of time slices after the second sample moment, and the second sample road section is a short-term sudden congestion road section;

and training to obtain a corresponding long-term prediction depth model based on third sample data, wherein the third sample data comprises long-term road condition characteristics of a third sample road section at a third sample moment and real traffic time values of time slices after the third sample moment, and the third sample road section is a long-term sudden congestion road section.

In one possible implementation, the relevant traffic characteristics of the target route include at least one of the following characteristics: the starting and ending point city, the starting time, the length of each road grade, the number of each type of intersection and the number of each type of turning action of the target route.

In one possible implementation, the method further includes:

and training to obtain the route-level prediction model based on fourth sample data, wherein the fourth sample data comprises the estimated passing time of the section level of the sample route, the relevant passing characteristics of the sample route and the real arrival time of the sample route.

In a second aspect, an embodiment of the present disclosure provides an ETA prediction model training method, including:

specifically, the ETA prediction model training method includes:

training to obtain a road section level prediction model corresponding to each road section type based on the sample data of each road section type;

calculating the predicted value of the passing time of each time slice of each type of sample road section in the sample route between the sample departure times based on the road section grade prediction model corresponding to each road section type;

determining the predicted entering time of each sample road section in the sample route based on the passing time predicted value of each sample road section in each time slice in the sample route;

accumulating the predicted passing time value of each sample road section in the corresponding target time slice to obtain the estimated passing time of the road section level of the sample route, wherein the target time slice corresponding to each sample road section is the time slice of the estimated entering time of each sample road section;

and training to obtain a route-level prediction model based on the estimated passing time of the section level of the sample route, the relevant passing characteristics of the sample route and the real arrival time of the sample route.

In a third aspect, an ETA prediction apparatus is provided in the embodiments of the present disclosure, including:

specifically, the ETA prediction apparatus includes:

the acquisition module is configured to acquire the road section type of each road section in the target route;

the calculation module is configured to calculate a predicted passing time value of each future time slice of each road section after the departure time based on a road section level prediction model corresponding to the road section type of each road section;

the determining module is configured to determine the predicted entering time of each road section based on the predicted passing time value of each road section in each future time slice;

the accumulation module is configured to accumulate the predicted passing time value of each road section in the corresponding target time slice to obtain the estimated passing time of the road section level of the target route, wherein the target time slice corresponding to each road section is a future time slice in which the estimated entering time of each road section is located;

a prediction module configured to predict an expected arrival time corresponding to the target route based on the estimated transit time at the link level of the target route and the relevant transit characteristics of the target route using a route level prediction model.

With reference to the third aspect, the present disclosure is in a first implementation manner of the third aspect, wherein the segment types include at least one of the following types: the method comprises the following steps of (1) low-frequency road sections, stable low-flow road sections, short-time sudden congestion road sections and long-time sudden congestion road sections; the computing module is configured to:

when the type of the road section is a stable low-flow road section, obtaining a predicted value of the passing time of the road section in each future time slice by using a linear model corresponding to a scene where the road section is located and based on the passing road condition characteristics of the road section;

when the type of the road section is a short-time sudden congestion road section, a short-time prediction depth model is used, and a passing time prediction value of the road section in each future time slice is obtained based on the short-time road condition characteristics of the road section;

and when the road section type of the road section is a long-term sudden congestion road section, obtaining the predicted value of the passing time of the road section in each future time slice by using the long-term prediction depth model and based on the long-term road condition characteristics of the road section.

In one possible implementation, the apparatus further includes:

a scene determining module configured to determine a scene in which the road segment is located based on at least one of the following characteristics of the road segment: the road section is located in the area, the time type of the departure time and the road grade of the road section.

In one possible implementation, the link type includes at least one of the following types under a primary link type: the method comprises the following steps of (1) low-frequency road sections, stable low-flow road sections, short-time emergent congestion road sections and long-time emergent congestion road sections, wherein the type of the first-stage road section comprises a common road section or a road section divided into directions at an intersection; the computing module is configured to:

when the type of the road section is a stable low-flow road section under the corresponding first-level road section type, obtaining a predicted value of the passing time of the road section in each time slice in the future by using a linear model corresponding to the scene where the road section is located based on the passing road condition characteristics of the road section under the corresponding first-level road section type, wherein the linear model is a model corresponding to the stable low-flow road section under the corresponding first-level road section type, and the passing road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time passing characteristics and historical average passing time sequences corresponding to the future time slices; the passing road condition characteristics of the road sections under the intersection direction-dividing road section types comprise at least one of the following characteristics: real-time traffic characteristics, historical average traffic time sequences corresponding to the future time slices, intersection types and steering action types;

when the type of the road section is a short-time sudden congestion road section under a corresponding first-level road section type, obtaining a predicted value of the passing time of the road section in each time slice in the future by using a short-term prediction depth model corresponding to the corresponding first-level road section type and based on short-term road condition characteristics of the road section under the corresponding first-level road section type, wherein the short-term road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time average passing time sequence in a short term, historical average passing time sequence corresponding to each future time slice and real-time traffic flow time sequence in the short term; the short-time road condition characteristics of the road section under the intersection diversion road section type comprise at least one of the following characteristics: real-time average passing time sequence in a short term, historical average passing time sequence corresponding to each future time slice, real-time traffic flow time sequence in the short term, intersection type and steering action type;

In one possible implementation, the apparatus further includes:

the first training module is configured to train and obtain a linear model corresponding to a target scene based on first sample data, wherein the first sample data comprises traffic road condition characteristics of a first sample section at a first sample moment and real traffic time values of time slices after the first sample moment, and the first sample section is a stable low-flow section in the target scene;

the second training module is configured to train to obtain a corresponding short-term prediction depth model based on second sample data, wherein the second sample data comprise short-term road condition characteristics of a second sample road section at a second sample moment and real traffic time values of time slices after the second sample moment, and the second sample road section is a short-term sudden congestion road section;

and the third training module is configured to train to obtain a corresponding long-term prediction depth model based on third sample data, wherein the third sample data comprises long-term road condition characteristics of a third sample section at a third sample time and real values of passing time of each time slice after the third sample time, and the third sample section is a long-term sudden congestion section.

In one possible implementation, the apparatus further includes:

the fourth training module is configured to train and obtain the route level prediction model based on fourth sample data, and the fourth sample data comprises the estimated passing time of the section level of the sample route, the relevant passing characteristics of the sample route and the real arrival time of the sample route.

In a fourth aspect, an ETA prediction model training apparatus is provided in the embodiments of the present disclosure.

Specifically, the ETA prediction model training device comprises:

the road section training module is configured to train to obtain a road section prediction model corresponding to each road section type based on the sample data of each road section type;

the middle calculation module is configured to calculate the predicted passing time value of each time slice of each type of sample road section between the sample departure times in the sample route based on the road section level prediction model corresponding to each road section type; determining the predicted entry time of each sample road section in the sample route based on the predicted value of the passing time of each sample road section in each time slice in the sample route; accumulating the predicted passing time value of each sample road section in the corresponding target time slice to obtain the estimated passing time of the road section level of the sample route, wherein the target time slice corresponding to each sample road section is the time slice of the estimated entering time of each sample road section;

and the route level training module is configured to train to obtain a route level prediction model based on the estimated passing time of the section level of the sample route, the relevant passing characteristics of the sample route and the real arrival time of the sample route.

In a fifth aspect, the disclosed embodiments provide an electronic device comprising a memory and a processor, wherein the memory is configured to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method of any aspect.

In a sixth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the method of any aspect.

In a seventh aspect, a computer program product is provided in the disclosed embodiments, comprising computer instructions that, when executed by a processor, implement the method steps of any aspect.

A seventh aspect of the present disclosure provides a navigation method, where a navigation route calculated at least based on a starting point, an ending point, and a road condition is obtained, an ETA of the navigation route is predicted and displayed, and navigation guidance is performed based on the navigation route, where the prediction of the ETA of the navigation route is implemented based on any one of the methods of the first aspect.

According to the technical scheme provided by the embodiment of the disclosure, different road section level prediction models can be used for predicting different road section types, a simple prediction scene is predicted by an efficient simple model, the problems of high complex model overhead and low calculation efficiency are solved, and the road section level predicted passing time of the target route considering the predicted entering time of each road section is taken as the input of the route level prediction model, so that the prediction model can accurately consider the future road condition of the user entering the corresponding road section at the future time instead of the current road condition, and the prediction accuracy is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

Other features, objects, and advantages of the present disclosure will become more apparent from the following detailed description of non-limiting embodiments when taken in conjunction with the accompanying drawings. In the drawings:

fig. 1 shows a flow diagram of an ETA prediction method according to an embodiment of the present disclosure.

FIG. 2 shows a flow diagram of an ETA predictive model training method according to an embodiment of the disclosure.

Fig. 3 illustrates a block diagram of an ETA prediction apparatus according to an embodiment of the present disclosure.

Fig. 4 shows a block diagram of an ETA prediction model training apparatus according to an embodiment of the present disclosure.

Fig. 5 shows an application diagram in a navigation application scenario according to an embodiment of the present disclosure.

Fig. 6 shows a block diagram of a server according to an embodiment of the present disclosure.

Fig. 7 shows a schematic structural diagram of a system of servers suitable for use to implement a method according to an embodiment of the present disclosure.

Detailed Description

Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. Furthermore, parts that are not relevant to the description of the exemplary embodiments have been omitted from the drawings for the sake of clarity.

In the present disclosure, it is to be understood that terms such as "including" or "having," etc., are intended to indicate the presence of the disclosed features, numbers, steps, behaviors, components, parts, or combinations thereof, and are not intended to preclude the possibility that one or more other features, numbers, steps, behaviors, components, parts, or combinations thereof may be present or added.

It should also be noted that the embodiments and features of the embodiments in the present disclosure may be combined with each other without conflict. The present disclosure will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

In the present disclosure, the acquisition of the user information or the user data is an operation that is authorized, confirmed, or actively selected by the user.

Fig. 1 shows a flow diagram of a method of predicted time of arrival prediction according to an embodiment of the present disclosure. As shown in fig. 1, the predicted arrival time prediction method includes the following steps S101 to S105:

in step S101, a link type of each link in the target route is acquired;

in step S102, calculating a predicted passing time value of each future time slice after the departure time of each road segment based on a road segment level prediction model corresponding to the road segment type of each road segment;

in step S103, determining the predicted entering time of each road section based on the predicted passing time value of each road section in each future time slice;

in step S104, accumulating the predicted passing time value of each road segment in the corresponding target time slice to obtain the estimated passing time of the road segment level of the target route, where the target time slice corresponding to each road segment is a future time slice in which the estimated entering time of each road segment is located;

in step S105, a route-level prediction model is used to predict a predicted arrival time corresponding to the target route based on the estimated transit time at the link level of the target route and the relevant transit characteristics of the target route.

As mentioned above, ETA (Estimated Time of Arrival) is an estimate of the length of a planned route trip. Typically, the navigation application will also use an ETA prediction algorithm to calculate for the user the ETA required for the user to reach the destination through the corresponding navigation route when planning the navigation route for the user. For navigation application with high daily life, the ETA prediction algorithm has considerable calling frequency, so that extremely high requirements are provided for the accuracy and the calculation efficiency of the ETA prediction algorithm. The current ETA prediction algorithm usually directly uses a prediction model to predict the current transit time of each road section on a navigation route, and the ETA of the navigation route is obtained through accumulation. The current passing time of different road sections in the prediction algorithm is obtained by prediction through a unified prediction model, and actually, the result can be rapidly predicted through a simple model for a plurality of road sections with stable passing time, so that the unified complex prediction model is used for different road sections in the existing ETA prediction algorithm, the calculation resource is wasted, the calculation efficiency is low, the current passing time of each road section is predicted in the existing model, the specific time of entering the road section is not considered, and the ETA predicted in such a way is not accurate.

In view of the above problems, the present disclosure provides an ETA prediction method, which can predict different road section types by using different road section-level prediction models, and can predict a simple prediction scene by using an efficient simple model, thereby solving the problems of high complex model overhead and low calculation efficiency, and taking the road section-level estimated transit time of a target route considering the estimated entry time of each road section as the input of the route-level prediction model, so that the prediction model can more accurately consider the future road condition of a user entering the corresponding road section at the future time rather than the current road condition, thereby improving the prediction accuracy.

In one possible implementation, the ETA prediction method may be applied to a computer, a computing device, an electronic device, a server cluster, and the like, which performs ETA prediction.

In one possible implementation, the target route may be a recommended navigation route from the departure point to the destination recommended by the navigation application for the user, or may be a navigation route from the departure point to the destination selected by the user.

In one possible implementation, a link refers to a basic unit of a road network, and is generally a small part of a road, and a link has only one entry and one exit, and the entry and exit at both ends of each link may be in topological connection with at least one other link. A target route may comprise a series of consecutive road segments.

In a possible implementation manner, the type of the road segment may be divided according to the stability degree of the passing time of the passing tool on the road segment, and after the passing road segment is divided, the road segment level prediction model corresponding to the type of the road segment may be trained for the type of the road segment. For example, if the transit time of each transit tool on the road segment is regular or stable in each time, the future transit time of such road segment is easy to predict, the road segment may be divided into a class of road segments, and a simple segment-level prediction model may be trained for the class of road segment, and if the transit time of each transit tool on the road segment is unstable in each time, and sudden congestion or the like is easy to occur, the future transit time of such road segment is not good to predict, the class of road segment may be divided into another class of road segments, and a complex segment-level prediction model may be trained for the class of road segment to accurately predict the future transit time of the class of road segment.

In one possible implementation, the link-level prediction model is used to predict the predicted transit time for each link at a future time slice after the departure time, where the time slice refers to a time slice, and for example, the duration of each time slice may be 5min, and then the future time slice after the departure time may be [0min-5min), [5min-10min) … … [ (5n-1) min-5n min), and so on.

In a possible implementation manner, still by using the above example, the road segments in the target route may be sorted in order from the departure location to the destination, the predicted entry time of the first road segment is the departure time, the predicted transit time of the first road segment in the time slice [0min-5min ] where the departure time is located is 10min, then the predicted entry time of the second road segment is 10min after the departure time, the predicted transit time of the second road segment in the time slice [10min-15min ] where the departure time is located is 8min, then the predicted entry time of the third road segment is … … after 18min after the departure time, and thus the predicted entry time of each road segment in the target route may be sequentially calculated.

In a possible implementation manner, the predicted passing time values of the sections at the predicted entering time are all accumulated together, so that the predicted passing time of the section level of the target route can be obtained, wherein the future time slice where the predicted entering time of each section is located can be determined as the target time slice, and the predicted passing time value of each section at the corresponding target time slice is the predicted passing time value of each section at the predicted entering time. Still taking the above example as an example, assume that the target route includes three road segments, i.e., a first road segment, a second road segment, and a third road segment, the predicted time of entry for the first road segment is the departure time, the target time slice corresponding to the first road section is a first time slice [0min-5min ] after the departure time, the predicted passing time value of the first road section in the target time slice [0min-5min ] is 10min, the target time slice corresponding to the second road section is [10min-15min ], the predicted passing time value of the second road section in the target time slice [10min-15min ] is 8min, the target time slice corresponding to the third road section is [15min-20min ], the predicted passing time value of the third road section in the target time slice [15min-20min ] is 10min, the estimated transit time at the link level for the target route is 28min +8min +10 min.

In one possible implementation, the input of the route-level prediction model is the estimated transit time of the link level of the target route and the relevant transit characteristics of the target route, and the output is the ETA corresponding to the target route, that is, the time expected to be needed from the departure place to the destination of the target route. And inputting the estimated passing time of the section level of the target route and the related passing characteristics of the target route into the route level prediction model, and executing the route level prediction model to obtain the ETA corresponding to the target route.

In one possible implementation, the relevant traffic characteristics of the target route may include at least one of the following characteristics: the starting and ending point city, the starting time, the length of each road grade, the number of each type of intersection and the number of each type of turning action of the target route. The starting and ending point city can be a city code of a city where the starting place is located and a city code of a city where the destination is located; the road grade may include various levels of an expressway, a national road, a provincial road, a county road, an urban road, a town road, and the like; the intersection types comprise predefined complex intersections and simple intersections, usually, a plurality of branches, overpasses, turntables and the like exist at the complex intersections, the intersection types in a road network can be predefined, and each intersection type in a target route can be directly obtained, so that the number of each type of intersection is obtained; the various steering actions comprise left steering, right steering, turning around and other various steering actions of the traffic tool when the traffic tool passes on the target route.

In this embodiment, different road section-level prediction models are configured for different road section types, and simple road section prediction can be carried out by using an efficient simple model for prediction, so that the problems of high complex model overhead and low calculation efficiency are solved, therefore, after the road section type of each road section in a target route is obtained, the passing time prediction value of each road section in future each time slice after departure time can be calculated according to the road section-level prediction model corresponding to the road section type, the predicted entering time of each road section is determined based on the passing time prediction value of each road section in each time slice, and the passing time prediction value of each road section at the predicted entering time is accumulated to obtain the predicted passing time of the road section level of the target route; the predicted passing time of the road section level obtained at the moment takes the predicted entering time of each road section into consideration, so that when a route level prediction model is used, and the ETA corresponding to the target route is predicted based on the predicted passing time of the road section level of the target route and the relevant passing characteristics of the target route, the predicted ETA takes the predicted entering time of each road section into consideration, so that the future road condition that a user enters the corresponding road section at the future time can be more accurately considered by the route level prediction model rather than the current road condition, and the prediction accuracy is improved.

In one possible implementation, the link types may further include at least one of the following types: the system comprises a low-frequency road section, a stable low-flow road section, a short-time sudden congestion road section and a long-time sudden congestion road section. The low-frequency road section refers to a road section which is occasionally passed by vehicles, and the low-frequency road section is basically smooth in each time slice and has no congestion; the stable low-flow road section refers to a road section which is less in congestion, regular and has no sudden congestion historically, such as a road section which is congested only in a peak period of a holiday; the short-time sudden congestion road section refers to a road section which has sudden congestion historically but has short congestion duration; the long-time sudden congestion road section refers to a road section in which sudden congestion with a long excessive duration historically occurs.

In a possible implementation manner, the step S102 of the ETA prediction method, namely, calculating the predicted passing time value of each future time slice of each road segment after the departure time based on the road segment level prediction model corresponding to the road segment type of each road segment, may include the following steps:

In this implementation manner, different road segment-level prediction models may be configured for different road segment types, for example, different road segment-level prediction models may be configured for a low-frequency road segment, a stable low-flow road segment, a short-time sudden congestion road segment, and a long-time sudden congestion road segment.

In this implementation, the simplest link-level prediction model, which may be a model for calculating the historical average transit time for the link, may be configured for the low frequency link. When the road section type of the road section is a low-frequency road section, vehicles occasionally pass through the low-frequency road section and are all smooth in all time slices, so that the historical average passing time of the road section can be used as the predicted passing time value of the road section in all future time slices.

In this implementation manner, a simpler linear model may be configured for the stable low-flow road segment, the stable low-flow road segment may be subjected to scene division according to scene characteristics, each scene trains one linear model, and the scene where the road segment is located may be determined based on at least one of the following characteristics of the road segment: the characteristics of the area where the road section is located (such as which city), the time type of departure time (such as early peak, flat peak or late peak, and the like), and the road grade of the road section. The linear models corresponding to different scenes can predict the predicted value of the passing time of the road section in each future time slice based on the passing road condition characteristics of the road section under different scenes.

In this implementation, the traffic condition characteristics of the road segment may include at least one of: real-time traffic characteristics and historical average traffic time sequences corresponding to future time slices; the real-time passing characteristic refers to the passing time of the road section at the current moment; assuming that the departure time is 9 am on monday, each time slice in the future is [9 am, 9 am 10min), [9 am 10min, 9 am 20min) … …, etc. on monday, and the historical average transit time sequence corresponding to each time slice in the future refers to the average value of the historical transit times of [9 am, 9 am 10min) on monday in history, the average value … … of the historical transit times of [9 am 10min, 9 am 20min) on monday in history, etc. And inputting the passing road condition characteristics of the road section into a linear model corresponding to the scene where the road section is located, so as to obtain the passing time predicted value of the road section output by the linear model in each future time slice.

In this implementation, a more complex short-term prediction depth model may be configured for the short-term sudden congestion road segment, and the short-term prediction depth model may predict the predicted transit time of the road segment in each future time slice based on the short-term road condition characteristics of the road segment, where the short-term road condition characteristics of the road segment include at least one of the following: the real-time average passing time sequence in a short term, the historical average passing time sequence corresponding to each future time slice and the real-time traffic flow time sequence in the short term. The real-time average transit time sequence in the short term may be an average transit time corresponding to a list of time sequences in a short time period, and for example, the real-time average transit time sequence in the short term may be a sequence of average transit times per minute in the past 1 hour of the current time. The short-term real-time traffic flow time series may be traffic flow information (such as inflow, outflow, and residence) corresponding to a list of time series on the road segment in a short time period, for example, the short-term real-time traffic flow time series may be a sequence of tool inflow, outflow, and residence per minute in the past 1 hour of the current time.

In this implementation manner, a complex long-term prediction depth model may be configured for the long-term sudden congestion road segment, and the long-term prediction depth model may predict a predicted value of the traffic time of the road segment in each future time slice based on the long-term road condition characteristics of the road segment, where the long-term road condition characteristics of the road segment include at least one of the following: real-time average traffic time sequence in a long term, historical average traffic time sequence corresponding to each future time slice, and real-time traffic flow time sequence in the long term; the real-time average transit time sequence in the long term may be an average transit time corresponding to a list of time sequences in a long period of time, and for example, the real-time average transit time sequence in the long term may be a sequence of average transit times per minute in the past 3 hours of the current time. The long-term real-time traffic flow time series may be traffic flow information (such as inflow, outflow, and remaining) corresponding to a list of time series on the road segment in a long period of time; for example, the long-term real-time traffic time series may be a series of traffic tool inflow, outflow, and dwell quantities per minute over the last 3 hours of the current time.

In the implementation mode, different road section level prediction models are configured for different road section types, decoupling of prediction effects of the different road section types is achieved, the simple prediction problem is solved through a simple and efficient model, the problem that in the prior art, complex models are used for large cost is solved, calculation efficiency is improved, finally, the prediction results of the road sections of various types can be mixed for use, and prediction accuracy is improved.

in the implementation mode, the intersection diversion road section means that the driving outlet of the road section is connected with other road sections in at least two directions, and when the passage tool passes through the intersection diversion road section, the speeds of different lanes in the intersection diversion road section are obviously different, for example, a left-turn lane is blocked, the passing speed is slow, the passing time is long, a straight lane is unblocked, the passing speed is fast, and the passing time is short. The ordinary road segment refers to a road segment other than the intersection diversion road segment.

In this implementation manner, the road segment types in the road network may include a low-frequency road segment, a stable low-flow road segment, a short-time sudden congestion road segment, and a long-time sudden congestion road segment in the common road segment types, and a low-frequency road segment, a stable low-flow road segment, a short-time sudden congestion road segment, and a long-time sudden congestion road segment in the intersection direction-dividing road segment types.

In this implementation, compared with the ordinary road section, the intersection diversion road section has more factors to be considered when predicting the future traffic time, for example, the type of the intersection, the steering action of the intersection, and other factors need to be considered, so that different road section level prediction models need to be set for the intersection diversion road section and the ordinary road section respectively. For example, different road section-level prediction models can be configured for a low-frequency road section, a stable low-flow road section, a short-time sudden congestion road section and a long-time sudden congestion road section under the type of a common road section, and a low-frequency road section, a stable low-flow road section, a short-time sudden congestion road section and a long-time sudden congestion road section under the type of a road section divided from an intersection.

In this implementation, when the link type of the link is the low-frequency link under the corresponding first-level link type, the simplest link-level prediction model may be configured for both the low-frequency links under the two first-level link types.

In this implementation, when the type of the road segment is a stable low-flow road segment under the corresponding type of the first-level road segment, because different types of the first-level road segment have different influence factors to be considered in prediction, different linear models need to be trained for different types of the first-level road segment, and in this implementation, one linear model needs to be trained for different scenes under different types of the first-level road segment. Using a linear model corresponding to a scene where the road section is located, and obtaining a predicted value of the passing time of the road section in each future time slice based on the passing road condition characteristics of the road section under the corresponding first-level road section type, wherein the linear model is a model corresponding to a stable low-flow road section under the corresponding first-level road section type, and the passing road condition characteristics of the road section under the common road section type include at least one of the following characteristics: real-time passing characteristics and historical average passing time sequences corresponding to the future time slices; the passing road condition characteristics of the road sections under the intersection direction-dividing road section types comprise at least one of the following characteristics: real-time traffic characteristics, historical average traffic time sequences corresponding to the future time slices, intersection types and steering action types; the intersection type comprises a complex intersection and a simple intersection, and the turning action type refers to a turning action type when the passing vehicle exits from an exit of the road section along the target route, and comprises actions of left turning, right turning, turning around and the like.

In the implementation mode, when the type of the road section is the short-time sudden congestion road section under the corresponding first-level road section type, a complex short-term prediction depth model can be configured for the short-time sudden congestion road section under the first-level road section type, influence factors needing to be considered in prediction of different first-level road section types are different, so that different short-term prediction depth models need to be trained for different first-level road section types, and the different short-term prediction depth models corresponding to different first-level road section types can predict the predicted value of the passing time of the road section in each future time slice based on the short-time road condition characteristics of the road section under different road section types. In this way, a short-term prediction depth model corresponding to a corresponding primary link type is used, and based on short-term road condition characteristics of the link in the corresponding primary link type, a predicted passing time value of the link in each time slice in the future is obtained, wherein the short-term road condition characteristics of the link in the common link type include at least one of the following: real-time average passing time sequence in a short term, historical average passing time sequence corresponding to each future time slice and real-time traffic flow time sequence in the short term; the short-time road condition characteristics of the road section under the intersection diversion road section type comprise at least one of the following characteristics: real-time average passing time sequence in a short term, historical average passing time sequence corresponding to each future time slice, real-time traffic flow time sequence in the short term, intersection type and steering action type;

in the implementation manner, when the type of the road section is the long-term sudden congestion road section under the corresponding first-level road section type, a complex long-term prediction depth model can be configured for the long-term sudden congestion road section under the first-level road section type, influence factors needing to be considered when the different first-level road section types are predicted are different, so that different long-term prediction depth models need to be trained for the different first-level road section types, and the different long-term prediction depth models corresponding to the different first-level road section types can predict the predicted value of the passing time of the road section in each future time slice based on the long-term road condition characteristics of the road section under the different road section types. The method comprises the steps of obtaining a predicted value of the passing time of a road section in each future time slice by using a long-term prediction depth model corresponding to a corresponding first-level road section type and based on the long-term road condition characteristics of the road section under the corresponding first-level road section type, wherein the long-term road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time average traffic time sequence in a long term, historical average traffic time sequence corresponding to each future time slice, and real-time traffic flow time sequence in the long term; the long-term road condition characteristics of the road section under the road section type of the intersection in different directions comprise at least one of the following characteristics: real-time average traffic time sequence in a long term, historical average traffic time sequence corresponding to each time slice in the future, real-time traffic flow time sequence in the long term, intersection type and steering action type.

In a possible implementation manner, the short-term prediction depth model or the long-term prediction depth model may include a first time-domain-gated Convolution layer (Temporal gate constraint), an image attention layer, a second time-domain-gated Convolution layer, a third time-domain-gated Convolution layer, and an MLP (multi layer Perceptron) layer, which are connected in sequence.

In this implementation, the model structure of the short-term prediction depth model or the long-term prediction depth model may be that an output end of the first time-domain-gated convolutional layer is connected to an input end of the graph attention layer, an output end of the graph attention layer is connected to an input end of the second time-domain-gated convolutional layer, an output end of the second time-domain-gated convolutional layer is connected to an input end of the third time-domain-gated convolutional layer, and an output end of the third time-domain-gated convolutional layer is connected to an input end of the MLP layer. Therefore, the short-term prediction depth model or the long-term prediction depth model can better extract the characteristics in the various time series, and further carry out accurate prediction.

In a possible implementation manner, the ETA prediction method may further configure different linear models for stable low-flow road segments located in different scenes under different primary road segment types, that is, the method may further include the following steps:

training to obtain a linear model corresponding to a target scene based on first sample data, wherein the first sample data comprises traffic road condition characteristics of a first sample road section at a first sample moment and traffic time real values of time slices after the first sample moment, and the first sample road section is a stable low-flow road section in the target scene.

In this implementation, the linear model corresponding to the stable low-flow road segment in different scenes may be trained by using the first sample data of the first sample road segment in different scenes at the first sample time. The traffic road condition characteristics in the first sample data can be input into an initial linear model, a traffic time predicted value of each time slice after the first sample moment output by the linear model is obtained, parameters in the linear model are adjusted according to errors between the traffic time predicted value of each time slice after the first sample moment and the traffic time true value of each time slice after the first sample moment until the errors are reduced to a certain degree, and then the linear model corresponding to the stable low-flow road section of the target scene is obtained through training.

In this implementation, the link types may include at least one of the following primary link types: when a common road section and an intersection are divided into direction road sections, different first-level road section types have different influence factors to be considered in prediction, so that different linear models need to be trained for different first-level road section types, and in the implementation mode, one linear model needs to be trained for different scenes under different first-level road section types. The method includes the steps that a corresponding linear model is obtained through training based on first sample data, wherein the first sample data comprise traffic road condition characteristics of a first sample road section under a corresponding first-level road section type at a first sample moment and traffic time real values of time slices after the first sample moment, and the first sample road section is a stable low-flow road section under the corresponding first-level road section type and is located in the target scene. The traffic road condition characteristics of the first sample section at the first sample moment under the common section type comprise at least one of the following characteristics: the passing characteristic at the first sample moment and the historical average passing time sequence of each time slice after the first sample moment; the passing road condition characteristics of the first sample road section at the first sample moment under the intersection direction-dividing road section types comprise at least one of the following characteristics: a traffic characteristic at a first sample time, a historical average traffic time sequence for each time slice after the first sample time, a crossing type, and a turning action type.

In a possible implementation manner, the ETA prediction method may further configure different short-term prediction depth models for the short-term sudden congestion road segments under different first-level road segment types, that is, the method may further include the following steps:

and training to obtain a corresponding short-term prediction depth model based on second sample data, wherein the second sample data comprises short-term road condition characteristics of the second sample road section at a second sample moment and real passing time values of time slices after the second sample moment, and the second sample road section is a short-term sudden congestion road section.

In this implementation, the short-term predictive depth model may be trained with second sample data for the second sample segment at the second sample time. The short-time road condition features in the second sample data can be input into an initial short-time prediction depth model to obtain a passing time prediction value of each time slice after a second sample time output by the short-time prediction depth model, parameters in the short-time prediction depth model are adjusted according to errors between the passing time prediction value of each time slice after the second sample time and the passing time true value of each time slice after the second sample time, and the short-time prediction depth model is obtained through training until the errors are reduced to a certain degree.

In this implementation, the road segment types may include at least one of the following primary road segment types: when a common road section and an intersection are divided into direction road sections, different first-level road section types have different influence factors to be considered in prediction, so that different short-term prediction depth models need to be trained for the different first-level road section types, and a short-term prediction depth model needs to be trained for the different first-level road section types in the implementation mode. The method comprises the steps of obtaining a short-term prediction depth model by training based on second sample data, wherein the second sample data comprises short-term road condition characteristics of a second sample road section of a corresponding primary road section type at a second sample moment and real traffic time values of time slices after the second sample moment, and the second sample road section is a short-term sudden congestion road section of the corresponding primary road section type. The short-time condition characteristics of the second sample segment of the normal segment type at the second sample time instant include at least one of: the average passing time sequence in a short period before the second sample time, the historical average passing time sequence corresponding to each time slice after the second sample time and the traffic flow time sequence in a short period before the second sample time; the short-time condition characteristics of the second sample segment of the intersection diversion segment type at the second sample time instant include at least one of: the average passing time sequence in a short period before the second sample time, the historical average passing time sequence corresponding to each time slice after the second sample time, the traffic flow time sequence in a short period before the second sample time, the intersection type and the steering action type.

In a possible implementation manner, the ETA prediction method may further configure different long-term prediction depth models for long-term sudden congestion road segments under different first-level road segment types, that is, the method may further include the following steps:

In this implementation, the corresponding long-term prediction depth model may be trained using third sample data of the third sample segment at the third sample time. The long-term road condition features in the third sample data can be input into an initial long-term prediction depth model to obtain a predicted value of the passing time of each time slice after a third sample time output by the long-term prediction depth model, parameters in the long-term prediction depth model are adjusted according to an error between the predicted value of the passing time of each time slice after the third sample time and a true value of the passing time of each time slice after the third sample time, and the long-term prediction depth model is obtained through training until the error is reduced to a certain degree.

In this implementation, the link types may include at least one of the following primary link types: when a common road section and an intersection are divided into direction road sections, different first-level road section types have different influence factors to be considered in prediction, so that different long-term prediction depth models need to be trained for different first-level road section types, and a long-term prediction depth model needs to be trained for different first-level road section types in the implementation mode. The method comprises the steps of obtaining a long-term prediction depth model by training based on third sample data, wherein the third sample data comprise long-term road condition characteristics of a third sample road section of a corresponding first-class road section type at a third sample moment and real traffic time values of time slices after the third sample moment, and the third sample road section is a long-term sudden congestion road section of the corresponding first-class road section type. The long-term road condition characteristics of the third sample road section of the common road section type at the third sample time comprise at least one of the following characteristics: the average passing time sequence in a long term before the third sample time, the historical average passing time sequence corresponding to each time slice after the third sample time and the traffic flow time sequence in a long term before the third sample time; the long-term road condition characteristics of the third sample road section of the intersection diversion road section type at the third sample time comprise at least one of the following characteristics: the average traffic time sequence in a long period before the third sample time, the historical average traffic time sequence corresponding to each time slice after the third sample time, the traffic flow time sequence in a long period before the third sample time, the intersection type and the steering action type.

In a possible implementation manner, the ETA prediction method may further include the following steps:

In this implementation, the relevant traffic characteristics of the sample route include at least one of: the starting and ending point cities, the starting and ending time, the lengths of all road levels, the number of all types of intersections and the number of all types of turning actions of the sample routes.

In the implementation manner, the estimated passing time of the section level of the sample route in the fourth sample data and the relevant passing characteristics of the sample route may be input into an initial route level prediction model to obtain the estimated arrival time of the sample route output by the route level prediction model, and parameters in the route level prediction model are adjusted according to an error between the estimated arrival time of the sample route and the real arrival time of the sample route until the error is reduced to a certain degree, so that the corresponding route level prediction model is obtained through training.

In this implementation, the route-level predictive model may be a DNN (Deep Neural Networks) model.

The implementation mode provides a double-layer prediction model architecture of road section level prediction-route level prediction, the deduction result predicted based on the road section level prediction model is further optimized by the route level prediction model to obtain the final predicted ETA, so that the route level model can be concentrated on prediction modeling of the passing time on various road sections, and the route level prediction model obtains higher accuracy by further adjusting the result.

The present disclosure also provides an ETA prediction model training method, and fig. 2 shows a flowchart of the ETA prediction model training method according to an embodiment of the present disclosure. As shown in fig. 2, the ETA prediction model training method includes the following steps S201 to S205:

in step S201, based on the sample data of each road segment type, a road segment level prediction model corresponding to each road segment type is obtained through training;

in step S202, based on the link-level prediction model corresponding to each link type, a predicted passing time value of each time slice after the sample departure time of each type of sample link in the sample route is calculated;

in step S203, determining an expected entry time of each sample road section in the sample route based on the predicted value of the transit time of each sample road section in each time slice in the sample route;

in step S204, accumulating the predicted passing time value of each sample road segment in the corresponding target time slice to obtain the estimated passing time of the road segment level of the sample route, where the target time slice corresponding to each sample road segment is the time slice of the estimated entry time of each sample road segment;

in step S205, a route-level prediction model is trained based on the estimated transit time of the link level of the sample route, the relevant transit features of the sample route, and the real arrival time of the sample route.

In one possible implementation, the ETA prediction model training method may be applied to a computer, a computing device, an electronic device, a server cluster, and the like that performs ETA prediction model training.

In one possible implementation, a link refers to a basic unit of a road network, and is generally a small part of a road, and a link has only one entry and one exit, and the entry and exit at both ends of each link may be in topological connection with at least one other link. A sample route may comprise a series of consecutive road segments.

In a possible implementation manner, the type of the road segment may be divided according to the stability degree of the transit time of the transit tool on the road segment, and after the transit road segment is divided, a road segment level prediction model corresponding to the type of the road segment may be trained for the type of the road segment. For example, if the transit time of each transit tool on the road segment is regular or stable in each time, the future transit time of such road segment is easy to predict, the road segment may be divided into a class of road segments, and a simple segment-level prediction model may be trained for the class of road segment, and if the transit time of each transit tool on the road segment is unstable in each time, and sudden congestion or the like is easy to occur, the future transit time of such road segment is not good to predict, the class of road segment may be divided into another class of road segments, and a complex segment-level prediction model may be trained for the class of road segment to accurately predict the future transit time of the class of road segment.

In one possible implementation, the link-level prediction model is used to predict the predicted transit time for each link in future time slices after the sample departure time of the sample route, where a time slice refers to a time slice, and for example, the duration of each time slice may be 5min, and then the time slices after the sample departure time may be [0min-5min), [5min-10min) … … [ (5n-1) min-5n min), and so on.

In a possible implementation manner, still by using the above example, the links in the sample route may be sorted in order from the departure point to the destination, the predicted entry time of the first sample link is the sample departure time, the predicted transit time of the first sample link in the time slice [0min-5min ] where the sample departure time is located is 10min, the predicted transit time of the second sample link after 10min from the sample departure time is 8min from the time slice [10min-15min ] where the sample departure time is located, and the predicted entry time of the third sample link is … … min after 18min from the sample departure time, so that the predicted entry time of each sample link in the sample route may be sequentially calculated.

In a possible implementation manner, the predicted passing time of the sample road sections at the predicted entering time is all accumulated together, so that the predicted passing time of the road sections of the sample route at the road section level can be obtained, wherein the future time slice of the predicted entering time of each sample road section at the predicted entering time can be determined as the target time slice, and the predicted passing time of each sample road section at the corresponding target time slice is the predicted passing time of each road section at the predicted entering time. Still taking the above example as an example, assume that the sample route includes three sample segments, i.e., a first segment, a second segment, and a third segment, the predicted entry time for the first segment is the departure time, the target time slice corresponding to the first road section is a first time slice [0min-5min ] after the departure time, the predicted passing time value of the first road section in the target time slice [0min-5min ] is 10min, the target time slice corresponding to the second road section is [10min-15min ], the predicted passing time value of the second road section in the target time slice [10min-15min ] is 8min, the target time slice corresponding to the third road section is [15min-20min ], the predicted passing time value of the third road section in the target time slice [15min-20min ] is 10min, the estimated transit time at the link level for the sample route is 10min +8min +10 min-28 min.

In one possible implementation, the relevant traffic characteristics of the sample route include at least one of the following characteristics: the starting and ending point cities, the starting and ending time, the lengths of all road levels, the number of all types of intersections and the number of all types of turning actions of the sample routes.

In a possible implementation manner, the estimated passing time of the section level of the sample route and the related passing characteristics of the sample route can be input into an initial route level prediction model to obtain the estimated arrival time of the sample route output by the route level prediction model, parameters in the route level prediction model are adjusted according to the error between the estimated arrival time of the sample route and the real arrival time of the sample route until the error is reduced to a certain degree, and then the corresponding route level prediction model is obtained through training.

In one possible implementation, the road segment types may further include at least one of the following types: the traffic congestion control method comprises the following steps of low-frequency road sections, stable low-flow road sections, short-time sudden congestion road sections and long-time sudden congestion road sections. The low-frequency road section refers to a road section which is occasionally passed by vehicles, and the low-frequency road section is basically smooth in each time slice and has no congestion; the stable low-flow road section refers to a road section which is less in congestion, regular and has no sudden congestion historically, such as a road section which is congested only in a peak period of a holiday; the short-time sudden congestion road section refers to a road section which has sudden congestion historically but has short congestion duration; the long-time sudden congestion road section refers to a road section with sudden congestion which has a long excessive duration historically.

In a possible implementation manner, different road segment-level prediction models may be trained for different road segment types, for example, different road segment-level prediction models may be trained for a low-frequency road segment, a stable low-flow road segment, a short-time congestion burst road segment, and a long-time congestion burst road segment, respectively.

In this implementation, the simplest link-level prediction model, which may be a model for calculating the historical average transit time for the link, may be configured for the low frequency link. When the type of the road section is the low-frequency road section, vehicles occasionally pass through the low-frequency road section and are smooth in all time slices, so that the historical average passing time of the road section can be used as the predicted passing time value of the road section in all future time slices.

In a possible implementation manner, the training method may also train different linear models for stable low-flow segments in different scenarios, that is, may further include the following steps:

In this implementation, the link types may include at least one of the following primary link types: when a common road section and an intersection are divided into direction road sections, different first-level road section types have different influence factors to be considered in prediction, so that different linear models need to be trained for different first-level road section types, and in the implementation mode, one linear model needs to be trained for different scenes under different first-level road section types. The method includes the steps that a corresponding linear model is obtained through training based on first sample data, wherein the first sample data include traffic road condition characteristics of a first sample section under a corresponding first-level section type at a first sample moment and traffic time real values of time slices after the first sample moment, and the first sample section is a stable low-flow section under the corresponding first-level section type and is located in the target scene. The passing road condition characteristics of the first sample road section at the first sample moment under the common road section type comprise at least one of the following characteristics: the traffic characteristic at the first sample time and the historical average traffic time sequence of each time slice after the first sample time; the passing road condition characteristics of the first sample road section at the first sample moment under the intersection direction-dividing road section types comprise at least one of the following characteristics: a traffic characteristic at a first sample time, a historical average traffic time sequence for each time slice after the first sample time, a crossing type, and a turning action type.

In a possible implementation manner, a more complex short-term prediction depth model may be trained for a short-term sudden congestion road segment, and influence factors to be considered in prediction of different primary road segment types are different, so that different short-term prediction depth models need to be trained for different primary road segment types, that is, the method may further include the following steps:

and training to obtain a corresponding short-term prediction depth model based on second sample data, wherein the second sample data comprises short-term road condition characteristics of the second sample road section at a second sample moment and real traffic time values of each time slice after the second sample moment, and the second sample road section is a short-term sudden congestion road section.

In this implementation, the road segment types may include at least one of the following primary road segment types: when a common road section and an intersection are divided into direction road sections, different first-level road section types have different influence factors to be considered in prediction, so that different short-term prediction depth models need to be trained for the different first-level road section types, and a short-term prediction depth model needs to be trained for the different first-level road section types in the implementation mode. The method comprises the steps of obtaining a short-term prediction depth model by training based on second sample data, wherein the second sample data comprises short-term road condition characteristics of a second sample road section of a corresponding primary road section type at a second sample moment and real traffic time values of time slices after the second sample moment, and the second sample road section is a short-term sudden congestion road section of the corresponding primary road section type. The short-time condition characteristics of the second sample segment of the normal segment type at the second sample time instant include at least one of: the average passing time sequence in a short period before the second sample time, the historical average passing time sequence corresponding to each time slice after the second sample time and the traffic flow time sequence in a short period before the second sample time; the short-time road condition characteristics of the second sample road segment of the intersection diversion road segment type at the second sample time moment comprise at least one of the following: the average passing time sequence in a short period before the second sample time, the historical average passing time sequence corresponding to each time slice after the second sample time, the traffic flow time sequence in a short period before the second sample time, the type of the intersection and the type of the steering action.

In a possible implementation manner, a more complex long-term prediction depth model may be trained for a long-term sudden congestion road segment under a first-level road segment type, and influence factors to be considered when predicting different first-level road segments are different, so the ETA prediction method may also train different long-term prediction depth models for long-term sudden congestion road segments under different first-level road segments, that is, may further include the following steps:

and training to obtain a corresponding long-term prediction depth model based on third sample data, wherein the third sample data comprises long-term road condition characteristics of a third sample road section at a third sample moment and real values of passing time of each time slice after the third sample moment, and the third sample road section is a long-term sudden congestion road section.

In this implementation, the corresponding long-term prediction depth model may be trained using third sample data of the third sample segment at the third sample time. The long-term road condition features in the third sample data can be input into an initial long-term prediction depth model to obtain a predicted passing time value of each time slice after a third sample time output by the long-term prediction depth model, parameters in the long-term prediction depth model are adjusted according to errors between the predicted passing time value of each time slice after the third sample time and a true passing time value of each time slice after the third sample time, and the long-term prediction depth model is obtained through training until the errors are reduced to a certain degree.

In this implementation, the road segment types may include at least one of the following primary road segment types: when a common road section and an intersection are divided into direction road sections, different first-level road section types have different influence factors to be considered in prediction, so that different long-term prediction depth models need to be trained for different first-level road section types, and a long-term prediction depth model needs to be trained for different first-level road section types in the implementation mode. The corresponding long-term prediction depth model can be obtained by training based on third sample data, wherein the third sample data comprises long-term road condition characteristics of a third sample road section of a corresponding first-class road section type at a third sample time and real values of passing time of each time slice after the third sample time, and the third sample road section is a long-term sudden congestion road section of the corresponding first-class road section type. The long-term road condition characteristics of the third sample road section of the common road section type at the third sample time comprise at least one of the following characteristics: the average passing time sequence in a long term before the third sample time, the historical average passing time sequence corresponding to each time slice after the third sample time and the traffic flow time sequence in a long term before the third sample time; the long-term road condition characteristics of the third sample road section of the intersection diversion road section type at the third sample time comprise at least one of the following characteristics: the average traffic time sequence in a long period before the third sample time, the historical average traffic time sequence corresponding to each time slice after the third sample time, the traffic flow time sequence in a long period before the third sample time, the intersection type and the steering action type.

In this implementation, the model structure of the short-term prediction depth model or the long-term prediction depth model may be that an output end of the first time-domain-gated convolutional layer is connected to an input end of the graph attention layer, an output end of the graph attention layer is connected to an input end of the second time-domain-gated convolutional layer, an output end of the second time-domain-gated convolutional layer is connected to an input end of the third time-domain-gated convolutional layer, and an output end of the third time-domain-gated convolutional layer is connected to an input end of the MLP layer. Thus, the short-term prediction depth model or the long-term prediction depth model can better extract the characteristics in the various time series, and further carry out accurate prediction.

In this embodiment, the explanation of the relevant contents in the ETA prediction model training method and the ETA prediction method is the same, and specific details can be referred to the description in the ETA prediction method, and are not described herein again.

Fig. 3 illustrates a block diagram of an ETA prediction apparatus according to an embodiment of the present disclosure. The apparatus may be implemented as part or all of an electronic device through software, hardware, or a combination of both. As shown in fig. 3, the ETA prediction apparatus includes:

an obtaining module 301 configured to obtain a link type of each link in a target route;

a calculating module 302 configured to calculate a predicted value of the passing time of each future time slice of each road segment after the departure time based on a road segment level prediction model corresponding to the road segment type of each road segment;

a determining module 303, configured to determine an expected entry time of each road segment based on the predicted value of the transit time of each road segment in each time slice in the future;

an accumulation module 304, configured to accumulate the predicted passing time value of each road segment in the corresponding target time slice to obtain the estimated passing time of the road segment level of the target route, where the target time slice corresponding to each road segment is a future time slice in which the estimated entering time of each road segment is located;

a prediction module 305 configured to predict a corresponding predicted arrival time of the target route based on the segment-level predicted transit time of the target route and the relevant transit characteristics of the target route using a route-level prediction model.

In one possible implementation, the type of the road segment includes at least one of the following types: a low-frequency road section, a stable low-flow road section, a short-time sudden congestion road section and a long-time sudden congestion road section; the calculation module 302 is configured to:

In one possible implementation, the apparatus further includes:

In one possible implementation, the link type includes at least one of the following types under a primary link type: the method comprises the following steps of (1) low-frequency road sections, stable low-flow road sections, short-time emergent congestion road sections and long-time emergent congestion road sections, wherein the type of the first-stage road section comprises a common road section or a road section divided into directions at an intersection; the calculation module 302 is configured to:

when the road section type of the road section is a stable low-flow road section under the corresponding first-level road section type, obtaining a passing time predicted value of the road section in each future time slice by using a linear model corresponding to a scene where the road section is located based on the passing road condition characteristics of the road section under the corresponding first-level road section type, wherein the linear model is a model corresponding to the stable low-flow road section under the corresponding first-level road section type, and the passing road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time passing characteristics and historical average passing time sequences corresponding to the future time slices; the passing road condition characteristics of the road sections under the road section types divided by the intersection comprise at least one of the following characteristics: real-time traffic characteristics, historical average traffic time sequences corresponding to the future time slices, intersection types and steering action types;

when the type of the road section is a short-time sudden congestion road section under the corresponding first-level road section type, obtaining a predicted value of the passing time of the road section in each future time slice by using a short-time prediction depth model corresponding to the corresponding first-level road section type and based on the short-time road condition characteristics of the road section under the corresponding first-level road section type, wherein the short-time road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time average passing time sequence in a short term, historical average passing time sequence corresponding to each future time slice and real-time traffic flow time sequence in the short term; the short-time road condition characteristics of the road section under the intersection diversion road section type comprise at least one of the following characteristics: real-time average passing time sequence in a short term, historical average passing time sequence corresponding to each future time slice, real-time traffic flow time sequence in the short term, intersection type and steering action type;

In a possible implementation manner, the short-term prediction depth model or the long-term prediction depth model includes a first time-domain gated convolutional layer, a graph attention layer, a second time-domain gated convolutional layer, a third time-domain gated convolutional layer, and a multi-layer perceptron MLP layer, which are sequentially connected.

In one possible implementation, the apparatus further includes:

the first training module is configured to train to obtain a linear model corresponding to a target scene based on first sample data, wherein the first sample data comprise traffic road condition characteristics of a first sample section at a first sample moment and traffic time real values of time slices after the first sample moment, and the first sample section is a stable low-flow section in the target scene;

and the third training module is configured to train to obtain a corresponding long-term prediction depth model based on third sample data, wherein the third sample data comprise long-term road condition characteristics of a third sample road section at a third sample moment and real traffic time values of time slices after the third sample moment, and the third sample road section is a long-term sudden congestion road section.

In one possible implementation, the relevant traffic characteristics of the target route include at least one of the following characteristics: the starting and ending city, the starting time, the length of each road grade, the number of each type of intersection and the number of each type of turning action of the target route.

In one possible implementation, the apparatus further includes:

In this embodiment, the ETA prediction apparatus corresponds to the ETA prediction method, and specific details can be referred to the description of the ETA prediction method, which is not described herein again.

Fig. 4 shows a block diagram of an ETA prediction model training apparatus according to an embodiment of the present disclosure. The apparatus may be implemented as part or all of an electronic device through software, hardware, or a combination of both. As shown in fig. 4, the ETA prediction model training apparatus 400 includes:

a road segment training module 401 configured to train to obtain a road segment prediction model corresponding to each road segment type based on sample data of each road segment type;

the middle calculation module 402 is configured to calculate a predicted passing time value of each time slice after the sample departure time of each type of sample road section in the sample route based on the road section-level prediction model corresponding to each road section type; determining the predicted entry time of each sample road section in the sample route based on the predicted value of the passing time of each sample road section in each time slice in the sample route; accumulating the predicted passing time value of each sample road section in the corresponding target time slice to obtain the estimated passing time of the road section level of the sample route, wherein the target time slice corresponding to each sample road section is the time slice of the estimated entering time of each sample road section;

a route-level training module 403 configured to train a route-level prediction model based on the estimated transit time at the link level of the sample route, the relevant transit features of the sample route, and the actual arrival time of the sample route.

In this embodiment, the ETA prediction model training apparatus corresponds to the ETA prediction model training method, and specific details may refer to the description of the ETA prediction model training method, which is not described herein again.

The embodiment of the disclosure also discloses a navigation service, wherein, based on the ETA prediction method, ETA of the navigated object using the navigation route is obtained, and based on the ETA of the navigated route, a navigation route selection service of a corresponding scene is provided for the navigated object. And the corresponding scene is one or a combination of multiple of long-distance AR navigation, overhead navigation or main and auxiliary road navigation.

The embodiment of the disclosure also discloses a navigation method, wherein the navigation route calculated at least based on the starting point, the end point and the road condition is obtained, the estimated arrival time ETA of the navigation route is predicted and displayed, and the navigation guidance is carried out based on the navigation route, wherein the prediction of the ETA of the navigation route is realized based on any ETA prediction method.

According to the embodiment of the disclosure, the predicted arrival time prediction method may be executed on a server, such as a cloud server, the server may predict the arrival time according to the predicted arrival time prediction method, and provide more accurate location services, such as a navigation service, a path planning service, and the like, for a mobile terminal, which may be a mobile phone, a pad, an IoT (Internet of Things) device, a vehicle-mounted terminal, and the like, and may display an electronic map. When the server navigates or plans the path for the mobile terminal, the server may obtain the predicted arrival time corresponding to the navigation route or each planned route based on the above method, and send the predicted arrival time to the mobile terminal, so that the mobile terminal displays the predicted arrival time of the corresponding route for the user, and the user may select a travel route or a travel time based on the predicted arrival time, and the like.

Fig. 5 shows an application diagram in a navigation application scenario according to an embodiment of the present disclosure. As shown in fig. 5, the mobile terminal and the server perform network communication by accessing a wireless network based on a communication standard, such as a mobile communication network of WiFi, 2G, 3G, 4G/LTE, 5G, or a combination thereof, when the user inputs a departure point and a destination for navigation, the mobile terminal may transmit the departure point and the destination input by the user to the server, the server may perform navigation route recommendation based on the departure point and the destination to obtain a plurality of candidate navigation routes from the departure point and the destination, then predict an expected arrival time corresponding to each candidate navigation route based on the above method, and select a predetermined number of recommended navigation routes from the plurality of candidate navigation routes based on the expected arrival time, the server may transmit each recommended navigation route and the expected arrival time corresponding to the recommended navigation route to the mobile terminal, the mobile terminal can receive and display each recommended navigation route and the corresponding predicted arrival time thereof, and the user can select a navigation route required by the user based on the predicted arrival time corresponding to each recommended navigation route, for example, the navigation route with the shortest predicted arrival time can be selected.

As shown in fig. 6, the server 600 includes a memory 601 and a processor 602, wherein the memory 601 is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor 602 to implement a method according to an embodiment of the disclosure.

Fig. 7 shows a system architecture diagram of a server suitable for use to implement the method according to an embodiment of the present disclosure.

As shown in fig. 7, the server system 700 includes a processing unit 701 that can execute various processes in the above-described embodiments according to a program stored in a Read Only Memory (ROM)702 or a program loaded from a storage section 708 into a Random Access Memory (RAM) 703. In the RAM 703, various programs and data necessary for the operation of the system 700 are also stored. The processing unit 701, the ROM 702, and the RAM 703 are connected to each other by a bus 704. An input/output (I/O) interface 705 is also connected to bus 704.

The following components are connected to the I/O interface 705: an input portion 706 including a keyboard, a mouse, and the like; an output section 707 including components such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage section 708 including a hard disk and the like; and a communication section 709 including a network interface card such as a LAN card, a modem, or the like. The communication section 709 performs communication processing via a network such as the internet. The driver 610 is also connected to the I/O interface 705 as needed. A removable medium 711 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 710 as necessary, so that a computer program read out therefrom is mounted into the storage section 708 as necessary. The processing unit 701 may be implemented as a CPU, a GPU, a TPU, an FPGA, an NPU, or other processing units.

In particular, the methods described above may be implemented as computer software programs, according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising computer instructions which, when executed by a processor, implement the method steps described above. In such an embodiment, the computer program product may be downloaded and installed from a network through the communication section 709, and/or installed from the removable medium 711.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The units or modules described in the embodiments of the present disclosure may be implemented by software or by programmable hardware. The units or modules described may also be provided in a processor, and the names of the units or modules do not in some cases constitute a limitation of the units or modules themselves.

As another aspect, the present disclosure also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the electronic device or the computer system in the above embodiments; or it may be a separate computer readable storage medium not incorporated into the device. The computer readable storage medium stores one or more programs for use by one or more processors in performing the methods described in the present disclosure.

The foregoing description is only exemplary of the preferred embodiments of the disclosure and is illustrative of the principles of the technology employed. It will be appreciated by those skilled in the art that the scope of the invention in the present disclosure is not limited to the specific combination of the above-mentioned features, but also encompasses other embodiments in which any combination of the above-mentioned features or their equivalents is possible without departing from the inventive concept. For example, the above features and (but not limited to) the features disclosed in this disclosure having similar functions are replaced with each other to form the technical solution.

Claims

1. A predicted time of arrival prediction method, comprising:

acquiring the section type of each section in the target route;

calculating a predicted value of the passing time of each road section in future each time slice after the departure time based on a road section grade prediction model corresponding to the road section type of each road section;

2. The method of claim 1, wherein the segment types include at least one of: the method comprises the following steps of (1) low-frequency road sections, stable low-flow road sections, short-time sudden congestion road sections and long-time sudden congestion road sections; the calculating the predicted value of the passing time of each future time slice of each road section after the departure time based on the road section level prediction model corresponding to the road section type of each road section comprises the following steps:

3. The method of claim 2, wherein the method further comprises:

determining the scene of the road section based on at least one of the following characteristics of the road section: the characteristics of the area where the road section is located, the time type of the departure time and the road grade of the road section.

4. The method of claim 2, wherein the segment type comprises at least one of the following types under a primary segment type: the system comprises a low-frequency road section, a stable low-flow road section, a short-time emergent congestion road section and a long-time emergent congestion road section, wherein the type of the first-stage road section comprises a common road section or a road section divided into directions at an intersection;

when the type of the road section is a stable low-flow road section under the corresponding first-level road section type, obtaining a predicted value of the passing time of the road section in each time slice in the future by using a linear model corresponding to the scene where the road section is located based on the passing road condition characteristics of the road section under the corresponding first-level road section type, wherein the linear model is a model corresponding to the stable low-flow road section under the corresponding first-level road section type, and the passing road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time passing characteristics and historical average passing time sequences corresponding to the future time slices; the passing road condition characteristics of the road sections under the road section types divided by the intersection comprise at least one of the following characteristics: real-time traffic characteristics, historical average traffic time sequences corresponding to the future time slices, intersection types and steering action types;

when the type of a road section of the road section is a long-term sudden congestion road section under a corresponding first-level road section type, obtaining a predicted value of the passing time of the road section in each future time slice by using a long-term prediction depth model corresponding to the corresponding first-level road section type based on the long-term road condition characteristics of the road section under the corresponding first-level road section type, wherein the long-term road condition characteristics of the road section under the common road section type comprise at least one of the following characteristics: real-time average traffic time sequence in a long term, historical average traffic time sequence corresponding to each future time slice, and real-time traffic flow time sequence in the long term; the long-term road condition characteristics of the road section under the road section type of the intersection in different directions comprise at least one of the following characteristics: real-time average traffic time sequence in a long term, historical average traffic time sequence corresponding to each time slice in the future, real-time traffic flow time sequence in the long term, intersection type and steering action type.

5. The method of claim 2, wherein the short-term prediction depth model or the long-term prediction depth model comprises a first time-domain-gated convolutional layer, a graph attention layer, a second time-domain-gated convolutional layer, a third time-domain-gated convolutional layer, and a multi-layer perceptron MLP layer connected in sequence.

6. The method of claim 2, wherein the method further comprises:

7. The method of claim 1, wherein the relevant traffic characteristics of the target route include at least one of: the starting and ending point city, the starting time, the length of each road grade, the number of each type of intersection and the number of each type of turning action of the target route.

8. The method of claim 1, wherein the method further comprises:

9. A method for training a prediction model of a predicted arrival time comprises the following steps:

determining the predicted entry time of each sample road section in the sample route based on the predicted value of the passing time of each sample road section in each time slice in the sample route;

10. An electronic device comprising a memory and a processor; wherein the memory is to store one or more computer instructions, wherein the one or more computer instructions are to be executed by the processor to implement the method steps of any of claims 1 to 9.

11. A readable storage medium having stored thereon computer instructions which, when executed by a processor, carry out the method steps of any one of claims 1 to 9.

12. A computer program product comprising computer instructions which, when executed by a processor, carry out the method steps of any of claims 1 to 9.

13. A navigation method, wherein a navigation route calculated at least based on a starting point, an end point and a road condition is obtained, an estimated arrival time of the navigation route is predicted and displayed, and navigation guidance is performed based on the navigation route, and the prediction of the estimated arrival time of the navigation route is realized based on any one of the methods of claims 1 to 9.